Use `async` locks in datastore, nix `asyncify`

Due to our use of non-async locks in the datastore, we have to asyncify i.e. spawn_blocking every time we acquire a transaction. This is bad for several reasons, including the non-zero overhead of spawn_blocking and the fact that procedures can acquire transactions from the database worker cores, whose single-core Tokio workers are not properly configured to run blocking threads. Instead, we should just use async-aware locks in our datastore, and await acquiring them.

Start by locating the datastore transaction-acquisition paths and the asyncify or spawn_blocking calls described in the issue. Trace how locks are acquired and identify the affected database-worker and procedure paths; done means those acquisitions use async-aware locks and await them without the existing blocking bridge.